Aiming at the problem of low fault diagnosis accuracy caused by the lack of fault samples for the rolling bearings of doubly fed wind turbines under normal conditions for a long time, an improved generative adversarial network fault diagnosis method based on expanding high-quality fault samples and using dual feature extraction is proposed. Firstly, a finite number of rolling bearing fault samples are expanded through a Wosselstein type generative adversarial network with maximum mean difference and penalty constraints; Then, based on the dual feature extraction model, the time-frequency converted temporal features and local features are extracted separately; Finally, the fault diagnosis of the rolling bearing balance data is completed through a classifier. The standard dataset and experimental results show that the proposed method improves fault diagnosis performance while lacking fault samples.